Europe’s AI Strategy Warns of Permanent Dependence Without a Compute Buildout

Europe’s Transformative AI Strategy frames the continent’s AI gap as a strategic-dependence problem, not simply a shortage of startups or venture funding. Its authors warn that without rapid action, Europe could become a long-term rule-taker—reliant on a small number of foreign governments and companies for the models, computing infrastructure and supply chains that shape economic activity, security and public life. 36

Why the authors see a risk of permanent marginalisation

The strategy says Europe is structurally unprepared for the arrival of more capable AI systems. Europe accounts for roughly 5% of global AI computing capacity, while European model developers’ annualised revenue is estimated at less than 2% of the combined revenue of two leading international competitors. 26

That concentration matters because access to AI is determined by more than software. It depends on chips, data centres, electricity connections, cloud infrastructure, frontier models and the ability of institutions to understand and respond to fast-moving technical change. The coalition’s warning is that once these dependencies are embedded in long-lived infrastructure and commercial contracts, they become much harder to unwind. 36

The headline target: 15% of global AI compute by 2030

The strategy recommends that the EU and its Member States aim to host 15% of global AI compute on EU soil by 2030—about three times Europe’s current share. 12

Its estimate is stark: reaching that goal would require around 45 GW of additional AI data-centre capacity and approximately €1.3 trillion in investment, with the bulk expected to come from private capital rather than governments funding the entire buildout themselves. 1

The recommendation is therefore also an implementation agenda. It calls for the conditions that make large infrastructure projects possible: quicker permitting, power availability, mechanisms to mobilise capital and reliable demand for computing capacity. 56

Supply-chain security: use Europe’s leverage before a crisis

A central proposal is a Member State Alliance for Supply Chain Security, coordinated with the European Commission. The strategy argues that European countries possess important AI supply-chain chokepoints but have not converted them into durable collective leverage. 18

The proposed alliance would seek more dependable access to chips, models and compute, while developing an escalation path for cases in which access is restricted. The strategy’s implementation outline includes a Commission liaison and inventory work in late 2026, followed by an assessment of conflicts and possible legal steps in 2027. 17

ASML is especially relevant to this argument because its lithography technology is a crucial part of the global semiconductor supply chain. The strategy’s broader point is not that Europe can independently control every component of the AI stack, but that it should coordinate the strategic assets it does have rather than negotiate from a position of fragmentation. 18

More than 40 recommendations beyond data centres

The package contains more than 40 recommendations. Its objectives extend beyond building capacity to ensuring that European researchers, firms and institutions can use advanced AI responsibly and reliably. 5

Key themes include:

• expanding access to frontier models and compute for European users;
• accelerating infrastructure, energy and capital mobilisation;
• building AI expertise and technical capacity inside public institutions;
• preparing for AI-enabled crises, including cyber-related risks; and
• maintaining democratic accountability as AI systems become more capable and widely deployed. 51418

This breadth is important: compute alone does not ensure influence. A region also needs capable buyers, skilled workers, public institutions that can evaluate the technology, and rules that can be enforced in practice.

Draghi’s alternative to foreign dependence—or a single European champion

Mario Draghi’s related argument is that Europe does not need to choose between productivity growth from AI and technological sovereignty. His proposed route is to retain meaningful control over European data and infrastructure while using large companies’ purchasing power to finance Europe-controlled data centres. 2224

The commercial model is pooled, multi-year demand. Companies including ASML, Capgemini and Amadeus have committed to purchasing Mistral’s European Compute Units, an arrangement intended to support 1 GW of capacity by 2030. 1922

The significance of this model is that forward purchasing commitments can make expensive infrastructure financeable before capacity is built. Rather than depending solely on public subsidy or existing hyperscalers, industrial customers commit to future compute use; that committed demand can underpin new capacity. 1930

Draghi’s argument is also explicitly anti-monopoly. The aim is not to trade foreign platform concentration for one protected European AI champion. A system of multiple providers and competing consortia could broaden choice while giving European customers more control over where their data and computing workloads reside. 19

Why the window is closing

The strategy’s urgency rests on the long lead times involved. Data centres need financing, equipment, sites and electricity; supply-chain arrangements and long-term enterprise contracts also take time to establish. Meanwhile, the strategy argues that other major powers are already building capacity and shaping the conditions of access. 36

Its core message is therefore practical: Europe’s AI future will be decided not only by regulation or research excellence, but by decisions on physical infrastructure, commercial demand and coordinated strategic action taken before dependency becomes the default. 11718